|
| ||
 |
I attended Ross Beveridge's colloquium on Automated Face Recognition at CSU and at Large. Dr. Beveridge presented some evaluation methods for face recognition algorithms as well as a number of face recognition algorithms. He also pointed out different face recognition applications requires different evaluation criteria (i.e. face recognition for security purpose cannot accept 2 people that have similar facial features, whereas facial recognition for matching can tolerate such errors). He presented the difficulty of face recognition with variation in light condition (shadow casted on side of face, different color tone based on color of light) and orientation (side, top view verses front view, distortion based on distance of view, etc). One interesting tidbit is that face recognition of physical changes over a period of time (aging, weight change) is much more difficult than face recognition over makeup, both for computers and human. While I don't know much about face recognition, popular forensics would lead one to believe that there are certain facial features somewhat immune (or at least have a predictable change pattern) to aging or weight change, such as jaw and cheek bone, etc. Last modified 4 December 2007 at 11:12 am by shumin |